: client : router : server
Fig. 4.3An evaluation topology.
download through the ECP-aware Interest retransmission control and the In-terest transmission control for congestion avoidance.
In the simulation we set up the adaptive video streaming scenario with 100 kinds of video content (a time length of each video content is 600 sec), com-posed of 300 video segments (a time length of each video segment is 2 sec).
Each server stores 3.7 Gbytes of video contents (MPD and 300 video seg-ments encoded in 20 bitrate levels (from 45 Kbps to 4.2 Mbps)) [63], and the RBA as the ABR algorithm is installed with the video player engine to each client. RBA determines the bitrate based on end-to-end throughput in the pre-vious download (described in section 2.2.1) every video segment download.
For comparison, we conduct 4 methods based on the aforementioned adap-tive video streaming scenario, adapadap-tive video streaming with LCE (RBA LCE), CASwECPN with LCE (CASwECPN RBA LCE), adaptive video streaming with Probcache (RBA Probcache), CASwECPN with Probcache (CASwECPN RBA Probcache).
4.2.1 Metrics for evaluation
To evaluate QoS, we evaluate throughput (Mbps) for both the video content download and the video segment download respectively.
To evaluate QoE, we use QoE-lin scoring [57–59]. QoE-lin allows the ob-jective assessment of QoE based on log information of an adaptive streaming application in a client. QoE-lin is a linear combination of three QoE metrics, a bitrate metric, a bitrate magnitude metric and a stall time metric.
To evaluate cache efficiency, we use the cache hit rate on each router and the details of received ECP Data at all the clients.
Table 4.1Average throughput.
Method Throughput per
video segment
Throughput per video content
RBA LCE 3.147 2.508
CASwECPN
RBA LCE 3.034 2.596
RBA
Probcache 3.119 2.486
CASwECPN RBA
Probcache 3.019 2.590
4.2.2 Evaluation in adaptive video streaming
Firstly, we evaluate the effect of CASwECPN on QoS metric, throughput in 4 methods. Table 4.1 and Figure 4.4 show the comparison of an average and a cumulative distribution function (CDF) of throughput (Mbps).
In Table 4.1, all the CASwECPN methods do not affect the average through-put metrics as compared to all the regular methods. This is because RBA enables the client to use network bandwidth more aggressively by increas-ing the bitrate (i.e., download traffic) as throughput increases. However, in Fig. 4.4a, we can see that throughput per video segment in all the CASwECPN methods do not become excessively small/large and proportion of around 2 Mbps is larger as compared to all the regular methods. Accordingly, all the CASwECPN methods enable the client to achieve a stable throughput. This is because the router avoids throughput reduction due to congestion by caching the traffics without transferring them to the congested network interface, and then transfers the cached traffic after congestion is mitigated. As a result, in Fig. 4.4b, all the CASwECPN methods are better in around throughput per video content is 3 Mbps as compared to all the regular methods.
(a)
(b)
Fig. 4.4 CDF for throughput. (a) per each video segment download, (b) per each video content download.
Table 4.2Average QoE metrics.
Method QoE-lin QoE
(bitrate)
QoE(bitrate magnitude)
QoE (stall time)
RBA LCE 179.787 729.860 257.170 292.904
CASwECPN
RBA LCE 348.640 715.900 219.343 147.918
RBA
Probcache 163.206 719.643 253.962 302.475
CASwECPN RBA Probcache
348.452 709.753 214.049 147.252
Secondly, we evaluate the effect of CASwECPN on QoE metrics, QoE-lin, QoE(bitrate), QoE(bitrate magnitude), and QoE(stall time) in 4 methods.
Table 4.2 and Figures 4.5-4.8 show the comparison of an average and a cu-mulative distribution function (CDF) of QoE metrics.
First, we evaluate the comprehensive QoE metric, QoE-lin. In Table 4.2, the CASwECPN RBA LCE method improves 94 % in the average QoE-lin as compared to the RBA LCE method. In addition, the CASwECPN RBA Prob-cache method also improves 114 % in the average QoE-lin as compared to the RBA Probcache methods. Besides, in Fig 4.5, all the CASwECPN meth-ods avoids excessively low QoE-lin as compared to all the regular methmeth-ods.
In addition, all the CASwECPN methods are not excessively high QoE-lin, and then QoE-lin becomes more fair for all the clients as compared to all the regular methods. Therefore, all the CASwECPN methods enable the client to improve the comprehensive QoE because the router mitigates QoE degrada-tion due to congesdegrada-tion (details are described in the following).
Second, we evaluate QoE(bitrate). In Table 4.2, all the CASwECPN methods do not affect the average QoE(bitrate). However, in Fig 4.6, all the CASwECPN methods avoid excessively high QoE(bitrate) larger than 600 as compared to all the regular methods. This is because CASwECPN enables the client to avoid an excessively high throughput causing congestion by caching the traffics during congestion (This behavior is shown in Fig. 4.4a). As a
result, all the CASwECPN methods avoid the excessively high bitrate.
Fig. 4.5CDF for QoE(linear combination).
Fig. 4.6CDF for QoE(bitrate).
Third, we evaluate QoE(bitrate magnitude). In Table 4.2, the CASwECPN RBA LCE method improves 15 % in the average QoE(bitrate magnitude) as compared to the RBA LCE method. In addition, the CASwECPN RBA Prob-cache method also improves 16 % in the average QoE(bitrate magnitude) as compared to the regular RBA Probcache method. Besides, in Fig 4.7, we can see that all the CASwECPN methods are smaller than 300 in QoE(bitrate magnitude) as compared to all the regular methods. Therefore, CASwECPN enables the router to mitigate congestion by avoiding excessive high bitrate (in Fig. 4.6) and bitrate reduction in RBA due to congestion. As a result, all the CASwECPN methods improve the QoE-lin.
Fourth, we evaluate QoE(stall time). In Table 4.2, all the CASwECPN methods improve 50 % in the average QoE(stall time) as compared to all the regular methods. Besides, in Fig. 4.8, we can see that all the CASwECPN methods are smaller than 400 in the QoE(stall time) although all the regular methods larger than 400 are 40 %. Therefore, CASwECPN enables the client to mitigate degradation of QoE-lin due to the stall time during congestion.
Consequently, RBA causes congestion because it selects the excessively high bitrate for available bandwidth based on implicit congestion estimation using the end-to-end throughput. In contrast, CASwECPN enables RBA to avoid the excessively high bitrate (in Fig. 4.6) because the router reduces the throughput quickly by caching data packets in its cache storage for congestion avoidance (in Fig. 4.4). Then, CASwECPN reduces the stall time by avoid-ing the excessively high bitrate (in Fig. 4.8), and enables RBA to mitigate the bitrate magnitude by adjusting the bitrate correctly according to explicit con-gestion state (inFig. 4.7). As a result, CASwECPN mitigates comprehensive QoE-lin degradation due to congestion (in Fig. 4.5).
Fig. 4.7CDF for QoE(bitrate magnitude).
Fig. 4.8CDF for QoE(stall time).
Table 4.3Average cache hit rate on each router.
Method Cache hit rate
RBA LCE 0.059
CASwECPN RBA LCE 0.073
RBA Probcache 0.017
CASwECPN RBA Probcache 0.071
Finally, we evaluate the effect of CASwECPN on cache efficiency metrics, cache hit rate on each router and the details of received ECP Data at all the clients.
Table 4.3 and Figure 4.9 show the comparison of an average and a cumu-lative distribution function (CDF) of cache hit rate per 10 seconds on each router in 4 methods.
In Fig. 4.9, all the methods are unable to use the router’s cache effectively in most of the periods because the video content size (3.7 Gbytes) is much larger than the router’s cache size (15 Mbytes). However, all the CASwECPN methods enable it to serve better the cache hit rate (from 0.4 to 0.6) than that of all the regular methods. This is because the clients surely request the explicitly cached contents according to ECPN and increases cache hit rate on the router storing the explicitly cached contents in congestion. As a result, CASwECPN shows a little improvement in the average cache hit rate on all the routers on a communication path (in Table 4.3) but the cashed content effectively works in the congestion situation. Besides, the RBA Probcache method excessively reduces cache hit rate because Probcache reduces the cache amount on the communication path as compared to LCE1. However, the CASwECPN RBA Probcache method improves the average cache hit rate as compared to the RBA Probcache method (in Table 4.3) because CASwECPN forces Probcache to cache the traffic during congestion and enables the client to request the cached content surely by ECPN after congestion is mitigated.
1Reduced cache hit rate of the regular Probcache method results in a little degradation of QoE-lin as compared to the regular LCE method in Fig. 4.5.
Fig. 4.9CDF for cache hit rate per 10 sec on each router.
Table 4.4Details of Received ECP Data.
Method RxRate Ratio of content sources in received ECP Data Downstream Congested R Upstream CASwECPN
RBA LCE 0.591 1.47×10
−3 0.999 1.53×10−6
CASwECPN RBA Probcache
0.595 1.41×10−3 0.999 5.42×10−6
Table 4.4 shows the details of received ECP Data at all the clients com-posed of reception rate (RxRate) of ECP Data ( N um of received ECP Data
N um of transmitted ECP Interest) and ratio of content sources, downstream from the router detecting congestion (the congested R), the congested R, upstream from the congested R, in all the received ECP Data.
In Table 4.4, RxRate is about 0.5 and the content source of ECP Data is almost 100 % of the congested router in all the CASwECPN methods. There-fore, in CASwECPN, the router also does not sufficiently mitigate the conges-tion at the first transfer of each ECP Data, and then explicitly caches each ECP Data again to mitigate congestion. As a result, the download time during con-gestion is delayed and the download throughput is reduced (in Fig. 4.4), and then RBA avoids excessively high bitrate (in Fig. 4.6). Thus, CASwECPN enables RBA to adjust the bitrate correctly according to congestion state.
Consequently, the conventional and implicit LCE and Probcache, which caches a content requested by a client for the other clients that may request the content subsequently, does not use the router’s cache effectively because routers are unable to keep caching the contents for the other users due to the very large video content size with comparing the cache size of the router. In contrast, ECPN enables the router that detects congestion to keep caching the content requested by a client until congestion is mitigated and surely deliv-ers the cached content to the client after congestion avoidance. As a result, CASwECPN improves cache hit rate of the routers notifying ECPN.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000 8000000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
FromCacheRate
Bitrate / Throughput (bps)
Video Segment Number
FromCacheRate Throughput Bitrate
Fig. 4.10 Time-series graph of video segment numbers with the bitrate, throughput, and FromCacheRate of each video segment in the RBA LCE method.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 1000000 2000000 3000000 4000000 5000000 6000000 7000000 8000000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
FromCacheRate
Bitrate / Throughput (bps)
Video Segment Number
FromCacheRate Throughput Bitrate
Fig. 4.11 Time-series graph of video segment numbers with the bitrate, throughput, and FromCacheRate of each video segment in the CASwECPN RBA LCE method.
To analyze the effect of cache control on the bitrate, we evaluate the time-series graph of video segment numbers that plots the bitrate, through-put, FromCacheRate of each video segment.
{F romCacheRate|0≤F romCacheRate≤1}represents the rate of data chunks were received from a router’s cache. FromCacheRate is defined in (4.1).
F romCacheRate= N um of data chunks received f rom cache
Sum of data chunks (4.1) Figure 4.10 and Figure 4.11 show the time-series graph of video segment numbers with the bitrate, throughput, and FromCacheRate of each video seg-ment in the RBA LCE method and the CASwECPN RBA LCE method.
In Fig. 4.10, we can see that RBA selects the excessively high bitrate for
the throughput after the throughput increases in the previous segment down-load with the high FromCacheRate (at segment number 4, 6, 11, 19). There-fore, since the implicit cache policy increases the throughput without consid-ering in-network congestion state, it causes the bitrate overestimation in RBA.
Then, the bitrate drops excessively, and the stall time experience is caused due to congestion, QoE degrades as a result.
On the other hand, in Fig. 4.11, we can see that CASwECPN reduces the throughput with the high FromCacheRate (at segment number 12, 13, 17, 18).
Therefore, CASwECPN reduces the throughput and avoids the bitrate over-estimation in RBA by downloading traffic during congestion with a delay for congestion avoidance. As a result, CASwECPN mitigates QoE degradation due to congestion as compared to the regular RBA methods (in Fig. 4.7 and Fig. 4.8).
Table 4.5 Average of client’s QoE-lin and router’s cache hit rate in CASwECPN RBA LCE method.
Num of Contents Cache Hit Rate QoE-lin
100 0.073 348.640
1000 0.073 351.481
10000 0.073 345.503
To investigate the cache efficiency in CASwECPN for the number of con-tents, we evaluate the cache hit rate and the QoE-lin in the CASwECPN RBA LCE method with increasing the number of video contents.
Table 4.5, Figure 4.12 and Figure 4.13 show the average and CDF of cache hit rate and QoE-lin in the CASwECPN RBA LCE method with 100, 1000, 10000 video contents.
In Table 4.5 and Fig. 4.12, we can see that the CASwECPN RBA LCE method achieves the same cache hit rate with all 100, 1000, 10000 video contents. Therefore, since CASwECPN enables the client to always request the explicitly cached contents, it works properly for the congestion state even if number of video contents is increased. As a result, in Table 4.5 and Fig.
4.13, the CASwECPN RBA LCE method achieves the same QoE-lin with all 100, 1000, 10000 video contents.
Fig. 4.12CDF of cache hit rate in the CASwECPN RBA LCE method with 100, 1000, 10000 contents.
Fig. 4.13 CDF of QoE-lin in the CASwECPN RBA LCE method with 100, 1000, 10000 contents.